The Generating Mechanisms of the August 17, 1999 I˙Zmit Bay (Turkey) Tsunami: Regional (Tectonic) and Local (Mass Instabilities) Causes

Marine Geology 225 (2006) 311–330 www.elsevier.com/locate/margeo

The generating mechanisms of the August 17, 1999 I˙zmit bay (Turkey) tsunami: Regional (tectonic) and local (mass instabilities) causes

Stefano Tinti a,*, Alberto Armigliato a, Anna Manucci a, Gianluca Pagnoni a, Filippo Zaniboni a, Ahmet Cevdet Yalc¸iner b, Yildiz Altinok c

a Universita` di Bologna, Dipartimento di Fisica, Settore di Geofisica, Viale Carlo Berti Pichat, 8, 40127 Bologna, Italy b Middle East Technical University, Department of Civil Engineering, Ocean Engineering Research Center, 06531, Ankara, Turkey c Istanbul University, Department of Geophysical Engineering, Division of Seismology, 34850, Avcilar, Istanbul, Turkey Received 23 November 2004; received in revised form 22 August 2005; accepted 2 September 2005

Abstract

The MW =7.4 earthquake that affected the northwestern part of Turkey on August 17, 1999, and in particular the gulf of I˙zmit, had dramatic consequences also as regards tsunami generation. The main cause of the earthquake was a dextral strike-slip rupture that took place along different segments of the western part of the North Anatolian Fault (WNAF). The rupture process involved not only a number of distinct strike-slip fault segments, but also dip-slip ancillary faults, connecting the main transcurrent segments. The general picture was further complicated by the occurrence of subsidence and liquefaction phenomena, especially along the coasts of the I˙zmit bay and in the Sapanca Lake. Tsunami effects were observed and measured during post-event surveys in several places along both the northern and the southern coasts of the bay. The run-up heights in most places were reported to lie in the interval 1–3 m: but in the small town of Deg˘irmendere, where a local slump occurred carrying underwater buildings and gardens of the waterfront sector, eyewitnesses reported water waves higher than 15 m. The purpose of this work is to investigate on the causes of the tsunami by means of numerical simulations of the water waves. We show that the tsunami was a complex event consisting at least of the combination of a regional event due to tectonic causes and of a local event associated with the mass failure. As to the first, we are able to demonstrate that the observed tsunami cannot be explained only in terms of the sea bottom dislocation produced by the main right-lateral dislocation, but that the prominent contribution comes from the displacement associated with the secondary shallow normal faults. Furthermore, the large waves and effects seen in Deg˘irmendere can be explained as the consequence of the slump. By means of a stability analysis based on an original method making use of the limit equilibrium concept, we show that the slump was highly stable before the earthquake and that it was triggered by seismic waves. Simulation of the tsunami induced by the slump was carried out by a two-step numerical code that computes the landslide motion first, and then the generated water wave propagation. It is shown that the computed local tsunami matches the experimental data. D 2005 Elsevier B.V. All rights reserved.

Keywords: 1999 Kocaeli earthquake; tsunami modeling; earthquake-induced tsunami; slump-induced tsunami

* Corresponding author. Tel.: +39 051 2095025; fax: +39 051 2095058. E-mail address: [email protected] (S. Tinti).

1. Introduction the last catastrophic occurrences of a long series of disastrous and lethal natural events that hit this region. 1.1. Tectonic setting and earthquakes Here earthquake generation is governed chiefly by three main tectonic structures: the East Anatolian Fault The Anatolian peninsula is one of the most seismi- (EAF), the North Anatolian Fault (NAF) and the Hel- cally active areas in the world. The 1999 Kocaeli lenic Arc (Fig. 1). The left-lateral EAF is circa 550 km earthquake and the associated I˙zmit bay tsunami are long and represents the border between the Anatolian

Fig. 1. Upper panel: Sketch of the Marmara Sea crossed by the Western section of the North Anatolian Fault (WNAF) system, which splits into three branches: northern, middle and southern branch. TB, Tekirdag˘ Basin; CMB, Central Marmara Basin; CB, Ç inarcik Basin; IB, I˙zmit Bay; KP, Kapidag˘ Peninsula; GP, Gelibolu Peninsula. Lower Panel: The WNAF is part of the complex fault pattern resulting from the Anatolian Plate dynamics. KJ, KarlVova Junction; NB, Niksar Basin; TEB, Tas¸ova-Erbaa Basin; HLB, Havza-Ladik Basin. S. Tinti et al. / Marine Geology 225 (2006) 311–330 313 block and the Arabian plate. The NAF is a dextral some tens of kilometers to the south, passing through system of faults forming the broad arc boundary be- the Manyas and the Uluabat lakes and through the tween the Eurasian plate and the Anatolian block. It area of Bursa (Gu¨len et al., 2002). Historical earth- extends for about 1500 km from the northern Aegean quake activity seems to be more intense on the south- Sea, through the Marmara Sea up to the easternmost ern than on the middle branch of the WNAF. Several part of Turkey where it connects to the EAF at the large earthquakes are known to have occurred here in Karliova triple junction. Its western section, denoted the last three centuries. The most recent sequence of hereafter as WNAF, splays into two branches (Fig. 1), major quakes manifested as an eastward migration of approximately in correspondence of the Mudurnu Val- rupture starting with the October 6, 1944 (MS 6.8) ley fault, east to Akyazi (Barka and Kadinsky-Cade, Edremit-Ayvalik earthquake, with epicentre of 1988; Barka, 1992; Gu¨len et al., 2002). The northern 26.568E, 39.488N (earthquake database of Bosphorus one starts from the North Aegean Sea, trending NE University, Kandilli Observatory, and Earthquake Re- along the gulf of Saros; then it crosses west–east the search Institute (KOERI), available at http://www. central and northern section of the Marmara Sea where koeri.boun.edu.tr/jeofizik/defaulteng.htm), and con- it broadens in a wide complex of strike-slip and dip-slip tinuing with the March 18, 1953 (MS 7.2) Yenice- faults; next, it penetrates in the˙ Izmit bay, whence it Go¨nen quake with epicentre of 27.368E, 39.998N continues towards the Sapanca Lake, further joining the (KOERI database), probably occurring on the middle main NAF system (Barka and Kadinsky-Cade, 1988). branch, and with the October, 06, 1964 (MS 6.9) A number of large historical earthquakes are associated Manyas-region shock: epicentre 26.238E, 40.308N with this branch. For example, fault rupture migrated (KOERI database). Hence the sequence affected the from east to west in the 18th century involving pro- overall sector from the Aegean Sea up to the Uluabat gressively all this system of faults from I˙zmit to S¸arkoÿ, Lake (Hubert-Ferrari et al., 2000; Ambraseys, 2002a; first with the 1719 (I˙zmit, MS 7.4), and then with the Barka et al., 2002). 1754 (eastern Marmara, MS 6.8) and with the May (eastern Marmara, MS 7.1) and August (Ganos, MS 1.2. Transtensional tectonics and pull-apart basins 7.4) 1766 shocks (Ambraseys and Finkel, 1995; Ambraseys, 2002a). Notice that westward migration Due to the collision between the Arabian and Eur- of seismicity has been identified in several sequences asian plates, the Anatolian plate is squeezed and pushed of earthquakes also along the main section of the NAF westward moving along the NAF and EAF. The tec- and may be related to the basic long-term tectonic tonic style is predominantly transcurrent along the NAF processes of inter-plate interaction going on here system, where slip occurs along a series of faults tens of (Barka, 1992; Armijo et al., 1999; Hubert-Ferrari et kilometers long with prevailing EW trend. But at the al., 2002). Later, major quakes that are attributed to junction of several of such segments, due to their segments of this branch occurred in 1878, in 1894 (MS relative offset and overstepping, extensional tectonics 7.3) and in 1912 (MS 7.4) (Ambraseys and Finkel, occurs with the consequent formation of a sequence of 1991; Ambraseys, 2002a). The most recent occurrences pull-apart basins, such as the prominent depressions of are the 1943 (MS 6.4) Hedek-Adapazari earthquake and the Niksar, of the Tas¸ova-Erbaa and of the Havza-Ladik the 1963 (MS 6.4) Ç inarcik earthquake. This is also the basins that one finds in the central portion of the NAF genetic place of the Kocaeli earthquake that is at the (Barka et al., 2000). In the WNAF, step-over depres- centre of our attention here, being the cause of the 1999 sions on land are often associated with lakes (e.g. the I˙zmit tsunami. Sapanca Lake in the northern branch of the WNAF, the The southern branch splits in turn into two further Iznik Lake in the middle strand and the Uluabat and the strands east to the Iznik Lake (Fig. 1), which makes Manyas lakes in the southern branch). Transtension and that the total number of the WNAF branches can be vertical movements are believed to be of foremost considered to be three: these further subsystems can importance especially on the northern branch of the be called the middle and the southern branch of the WNAF. Indeed the genesis of the Marmara Sea itself WNAF. The middle runs NE-ward from the Aegean is attributed to pull-apart processes (Hubert-Ferrari et Sea, south of the Dardanelles Straits, and intersects the al., 2002). More specifically, overstepping may be ad- southern coast of the Marmara Sea around BandVrma. vocated to explain the formation of the three main inner From here it parallels the coastline up to the Gemlik basins characterising the Marmara Sea (see Armijo et bay, south of the Armutlu peninsula. The WNAF al., 2002; Flerit et al., 2003) that from west to east are southern branch trends almost parallel to the middle, the Tekirdag˘ basin, the Central basin and the Ç inarcik 314 S. Tinti et al. / Marine Geology 225 (2006) 311–330 basin, separated from one another by bathymetric highs shallow being at most 40 m deep (Lettis et al., 2002; (Fig. 1; see Le Pichon et al., 2001). And analogously, Kus¸çu et al., 2002). processes associated with fault segment step-overs may be invoked also to interpret the morphological structure 1.3. Tsunamis of the 50 km long and 10 km wide I˙zmit bay, where three distinct sub-basins can be identified (Fig. 2): the A long history of tsunami occurrences is associated western basin, which is the deepest one, and opens into with the earthquakes in the coastal and submarine seg- the Marmara Sea, the Karamu¨rsel basin (central basin) ments of the WNAF, and especially of the northern with maximum depth around 200 m, and the eastern strand. As many as circa 100 tsunami events are basin, hereafter named the I˙zmit basin that is very reported in tsunami catalogues of Turkey, starting

Fig. 2. Fault system in the eastern part of the Marmara Sea that is assumed to be involved in the tsunami generation (lower panel). It is composed by strike-slip faults (i.e. the Marmara, MF, the Hersek, HF, the Go¨lcu¨k, GF, and the Sapanca fault, SF), and by normal faults (i.e. the Karamu¨rsel faults, KARF1, KARF2, KARF3 and KARF4, the Kavakli fault (KAVF) and the I˙zmit faults (IF1 and IF2)). Fault parameters are listed in Table 2. Names and bathymetry of the basins forming the I˙zmit bay are given in the upper map. S. Tinti et al. / Marine Geology 225 (2006) 311–330 315 from the Minoan Santorini explosion tsunami in the region from the Eften Lake, east of Karadere, to the Aegean Sea until the recent 1999 Kocaeli-earthquake whole I˙zmit bay. The areas with the largest toll of tsunami that is treated here (see Altinok and Ersoy, destruction and damage were the district of Adapazari 2000; Papadopoulos, 2001; see also Ambraseys, and the belt around the eastern and southeastern coast 1962; Kuran and Yalçiner, 1993). Practically, according of the I˙zmit bay, but the earthquake was damaging also to historical records, all the Turkish coasts were affect- west of I˙stanbul where severe effects were observed in ed: the Black Sea coastline to the north (Yalçiner et al., the coastal towns of Buÿu¨kçekmece and Ku¨çu¨kçek- 2004), the Aegean coast to the west and the eastern mece in the district of Avcilar. The earthquake was Mediterranean coasts (Yalçiner et al., 2001a,b), but one lethal, killing more than 17,000 people. The number of the largest concentration of cases is found in the of injuries exceeded 43,000, and as many as about Marmara Sea and in the˙ Izmit bay. Historical informa- 250,000 people remained homeless, implying property tion on tsunamis in Turkey, as in all the other European losses in the range of $6.5 billion (USGS Report, countries and in many other countries in the world, is 2000). The magnitude was estimated to be MW =7.4 generally insufficient for a statistical analysis of the by Kandilli Observatory and the epicentre was located tsunami hazard and for a proper characterisation of at 40.88N and 308E (with hypocentre around 17 km the tsunami effects. Reports are mostly sporadic and depth) near the eastern end of the I˙zmit bay (Barka et detailed descriptions are available only for a restricted al., 2002). Shaking lasted about 45 s and was recorded number of cases. Moreover, the reliability of the his- by several strong motion sensors placed around and torical sources is not always sound. In a recent study on very near to the source (managed by the Earthquake the Marmara Sea tsunamis, Ambraseys (2002b) con- Research Department of General Directory of Disaster cluded that about 50% of the cases found in the liter- Affairs of Turkey). The highest recorded ground accel- ature are to be ruled out as spurious events. eration was 0.45 g at the station of Sakarya, to the east Nonetheless, this region remains one of the European of the Sapanca Lake. The earthquake ruptured at least areas with the highest tsunami activity, and most vul- five segments of the northern branch of the NAF, each nerable to tsunami attacks. Interest for the evaluation of with length in the range of 20–35 km that from east to the tsunami risk in the whole region of the Marmara west are the Karadere segment, the Sapanca-Akyazi Sea has increased after the occurrence of the 1999 segment (to the east of the Sapanca Lake), the Sapanca Kocaeli earthquake and tsunami, and after the specula- segment (to the west of the lake), the Go¨lcu¨k segment tion based on stress-transfer theory that its occurrence and the Hersek segment (Barka et al., 2002; Lettis et has highly favoured the condition for the next WNAF al., 2002). The tsunamigenic fault segments are rupture to take place in the submarine section of the sketched in Fig. 2. Aftershock distribution and SAR northern strand (Hubert-Ferrari et al., 2000; Parsons et data analysis possibly suggest that rupture took place al., 2000; Parsons, 2004). A possible future earthquake also on a further segment in the Marmara Sea to the occurring in the Marmara Sea has a direct tsunamigenic east of the I˙zmit bay entrance (Gu¨len et al., 2002; Ç akir potential, and, furthermore, may set in motion subma- et al., 2003). The areas connecting the fault segments rine masses (landslides or slumps), with additional have been recognised as distinct right-releasing step- possibility to produce tsunamis. Signatures of past over regions, whose width goes from 1 to 2 km for the mass failures have been identified in several places by Sapanca Lake and the Go¨lcu¨k step-overs up to 4–5 km specific and very detailed marine geological surveys in for the Karamu¨rsel step-over (Lettis et al., 2002). Slip several places of the sea of Marmara (see e.g. Altinok et distribution deduced from seismological records and al., 1999; Alpar and Yaltirak, 2002; Yu¨ksel et al., 2002), from field surveys of the surface breaks shows that and scenarios of tsunamis induced by earthquake-trig- slip was highly heterogeneous along the fault and that gered underwater slides have been analysed by Yalçiner the maximum values of dextral displacement have to be et al. (2002), with the conclusion that risk associated assigned to the Sapanca-Akyazi segment (5–6 m) and with tsunami occurrence is a very serious issue espe- to the Go¨lcu¨k segment (4–5 m) (Barka et al., 2002; cially in the I˙stanbul area. Lettis et al., 2002), a picture that is basically further confirmed by inversion of GPS and satellite data (Feigl 2. The Kocaeli earthquake and its genetic fault et al., 2002; see also Delouis et al., 2002, whose study system identifies a third area of increased slip in the westernmost sector of the I˙zmit bay) as well as by combining The Kocaeli earthquake struck in the night of Au- tectonic observations and SAR data (Ç akir et al., gust 17, 1999 at 3:02 AM and affected a very vast 2003). 316 S. Tinti et al. / Marine Geology 225 (2006) 311–330

The multiple-rupture main shock was dominated 3. Tsunami observations by the right-lateral strike-slip component in agreement with the dominant tectonic style of the region. The 1999 Kocaeli earthquake caused a tsunami that Consistently, observed ground movement was typical- was observed in the entire I˙zmit bay. Unfortunately no ly horizontal. However, also vertical ground move- tide-gauge records are available for the tsunami either ments were identified in several places in a narrow in the source region or in the Marmara Sea. Post-event belt along the main fault-segments zone. Coastal field surveys and interview of eyewitnesses conducted subsidence was observed in the southern coast of soon after the tsunami occurrence by several teams the Sapanca Lake and in the southeastern coast of enabled to determine the coastal areas of major wave the I˙zmit bay, and caused extensive flooding and impact and to quantify the run-up heights and the sea permanent retreat of the shoreline by tens to hundreds penetration lengths in all the affected places. The tsu- of meters. The observed vertical displacements can be nami was observed soon after the main shock. Almost accounted for only partially by the strike-slip source in all places the tsunami was seen to cause the sea to model, and require a different explanation, namely withdraw first and then to overcome the usual shoreline liquefaction and normal faulting. Ground sinking and inundate land. The observed tsunami run-up due to widespread liquefaction phenomena was ascer- heights were modest and typically in the range of 1–3 tained to have occurred in Adapazari where several m. The tsunami was damaging, but not catastrophic. In hundreds of buildings were affected (USGS Report, some places it was strong enough to carry boats and 2000). Liquefaction was also observed in Sapanca as vessels on land. One example comes from Yenikoÿ, east well as in the delta fan east of Go¨lcu¨k (Sucuog˘lu et of Kavakli, where a large floating platform, 60 m long al., 2000). and 12 m wide that was used as a pier, was moved and Ancillary normal faults with dominant vertical com- transported about 100 m on land over a grassy stretch of ponent in addition to a smaller right-lateral slip were coast (Altinok et al., 2001; Rothaus et al., 2004). A mapped in several places. The most relevant one was detailed description of the tsunami observations can be seen east of Go¨lcu¨k in the locality known as Kavakli, found in the works by Yalçiner et al. (2000, 2001a,b,c) and was the main cause of the subsidence that affected and by Altinok et al. (2001), and has been integrated in Go¨lcu¨k (Rothaus et al., 2004). The fault, trending NW– a recent study by Rothaus et al. (2004) that focuses on SE, can be traced for a 3.5 km inland, intersects the the effects in the southern coast of the I˙zmit bay. Table coastline, and probably continues offshore. The scarp 1 summarises the observations on first arrival, water shows a vertical offset of about 2 m and a typical penetration and run-up heights that can be found in the dextral offset of 0.40–0.60 m that were measured by aforementioned papers, and complements the data set several surveying teams (Barka et al., 2002; Lettis et al., by adding new observations, so far unpublished that 2002), one of which was a joint Italian-Turkish (IT) were collected during the IT post-event field survey. team and included four of the authors of the present The tsunami was exceptionally violent only in the paper (namely Tinti, Armigliato, Yalçiner and Altinok). town of Deg˘irmendere. Here measured wave run-ups The role of surface normal faulting was probably minor exceeded 4 m, but several eyewitnesses state that the in determining the major catastrophic ground shaking wave reached the second floor of the buildings in the that is to be attributed more appropriately to the rup- waterfront (IT), and some other estimate a height of tures along the long dextral fault segments. But the about 15 m (Rothaus et al., 2004). Ships moored in the point of this paper is that these faults were essential port of Deg˘irmendere were severely affected by the in the genesis of the tsunami. Seismic profiles taken waves and seamen told that the attacking wave was after the occurrence of the Kocaeli earthquake show the probably larger than 10 m (Altinok et al., 2001). Per- clear signature of the main fault zone that disturbed the haps the same large wave struck ships moored in the underwater sediments (Alpar, 1999). However, beyond next town of Kavakli, east of Go¨lcu¨k: here the tsunami the main fault system, offshore normal faults were came over a ship reaching a height of more than 15 m, identified by means of bathymetric and shallow seismic according to witnesses rough evaluation (Rothaus et al., data in the central (Karamu¨rsel) and in the eastern parts 2004). The anomalous violence of the tsunami in the of the I˙zmit bay (S¸engo¨r et al., 1999; Alpar and Yal- limited area around Deg˘irmendere is the consequence tirak, 2002; Lettis et al., 2001, 2002), and their trend of an additional local tsunamigenic source that was and position is consistent with the vertical dynamics identified in a slump. A stretch of coast about 300 m that is associated with the formation of these step-over long (parallel to shoreline) and 75 m wide (perpendic- basins. ular to shoreline), where a recreational area called S. Tinti et al. / Marine Geology 225 (2006) 311–330 317

Table 1 induced by the coastal mass failure is the main aim of Post-tsunami field survey observations this work and will be tackled with the appropriate Locality Locality Survey First Penetration Observed attention in the following. codea arrival length run-up (m) height (m) 4. Simulations of the earthquake-induced tsunami DarVca 1 AL 4 N1 Eskihisar 2 AL 15 N1 Tsunami simulations are based on a finite-element Dilovasi 3 AL 2 model solving the non-linear shallow water equations Tavs¸ancVl4 AL 25–30 1.5 that has been developed specifically for tsunami anal- IT 25 1.7–2.5 ysis by the University of Bologna, Italy, and has been Hereke 5 AL 30 1.8 extensively used to study near-field propagation of IT 2.25–2.9 S¸irinyali 6 AL 15 N2 tsunamis generated by near-shore earthquakes (Tinti Kirazliyali 7 AL 2.5 et al., 1994; among the applications, see Piatanesi and YarVmca 8 AL N60 Tinti, 1998, 2002; Tinti and Armigliato, 2003). The IT 3.2 I˙zmit bay and the easternmost part of the Marmara Ko¨rfez AL 100 Sea facing the bay have been covered by a grid con- Tupras 9 AL 20 2.5 (Kiler Point) sisting of 11,006 nodes and 20,197 triangles, having a Derince Port 10 AL 2–2.5 space resolution variable from place to place (coarser in Ç ene Suyu 11 AL 60 2? deep water and finer in shallow coastal areas). The grid I˙zmit, 12 AL 25 that is portrayed in Fig. 3 was created through a mesh Marina building code optimised for tsunami simulations that is Seymen 13 AL 50 (Yenikoÿ) suited to basins with complex bathymetries and irregu- IT 50 2 lar coastlines and that tends to produce triangles with RO N200 the highest index of isotropy (typically equilateral) and Kavakli 14 AL 300? a uniform crossing time (Tinti and Bortolucci, 1999). b Go¨lcu¨k 15 RO 15–20 The coastal boundary is delimited by vertical walls Deg˘irmendere 16 AL 35? 2.5 IT 2.4 causing total reflection of the water waves. Technically, RO 80 4.4 (15b) this means that the tsunami model does not compute the Halidere 17 AL 60 0.8 tsunami run-up heights, but only the maximum water IT 47–92 1.5 elevation (MWE) at the vertical coast. However, this is N Ulasli 18 AL 52 not a significant limitation for the purpose of this paper, Eregli- 19 AL + 4 1.25 Guzelyali since MWEs are known to be a reasonable approxima- Defne 20 AL 4 1.5 tion of run-up heights for most coastal morphologies. Mahallesi The depth data set has been assembled by digitising Karamu¨rsel 21 AL nautical maps of the Marmara Sea (1:300,000) and of Kaytazdere 22 AL 15? the I˙zmit bay (1:50,000) published by the Hydrographic Hersek 23 AL N30 RO 15 and Oceanographic Service of Turkey and by digitising Havuzdere AL maps from Lettis et al. (2002) that are based on recent Topçular 24 AL 2? data gathered by the Turkish Navy and by Gu¨neysu Offshore AL 30? 30? (1999). This bathymetry data set has been further inte- Topçular grated with depth data collected in the course of the AL: survey by Altinok et al. (2001); IT: survey by Italian-Turkish post-tsunami survey by the IT team. Fig. 2 provides a team composed by Altinok Y., Armigliato A., Bortolucci E., Guidi C., map with the bathymetry and Fig. 3 shows the grid and Tinti S., Yalçiner A.C.; RO: survey by Rothaus et al. (2004). a Localities are numbered as in Fig. 3. localities where tsunami observations were made and/or b Not run-up height value, but wave height offshore reported by tsunami time histories and maximum water elevations eyewitnesses. have been calculated. The faults used to generate the numerical tsunami Ç inarlik Park was located, slid into the sea, carrying a are listed in Table 2. As regards the main strike-slip multi-storing hotel, owned by the municipality, a cafe faults, we consider only those fault segments that are and a restaurant and 14 park trees (Arel and Kiper, found close enough to the water basin to cause defor- 2000; Altinok et al., 2001; Cetin et al., 2004). Modeling mation of the sea bottom. Therefore neither the Kara- of both the tectonic tsunami and the local tsunami dere segment nor the Sapanca-Akyazi segment are 318 S. Tinti et al. / Marine Geology 225 (2006) 311–330

Fig. 3. Grid used for the numerical simulation of the earthquake-induced tsunami, consisting of 11,006 nodes and of 20,197 triangles. Numbers refer to places where tsunami observations of the post-event surveys were made (see Table 1): (1) Darica, (2) Eskihisar, (3) Dilovasi, (4) Tavs¸ancil, (5) Hereke, (6) Sirinyali, (7) Kirazliyali, (8) Yarimca, (9) Tupras, (10) Derince, (11) Cene Suyu, (12) I˙zmit, (13) Seymen, (14) Kavakli, (15) Go¨lcu¨k, (16) Deg˘irmendere, (17) Halidere, (18) Ulas¸li, (19) Eregli, (20) Defne, (21) Karamu¨rsel, (22) Kaytazdere, (23) Hersek, (24) Topc¸ular. taken into account. The faults supposed to contribute to most rupturing segment of the Kocaeli earthquake, the tsunami are from east to west the Sapanca fault although this point of view is not unanimously agreed (SF), the Go¨lcu¨k fault (GF) and the Hersek fault (HF). upon. Fault parameters are derived from the literature, We also add the Marmara fault (MF) that is totally which is properly acknowledged in the caption of Table submarine and that, according to some studies (Gu¨len 2. Some of these faults are subdivided in a number of et al., 2002; C¸ akir et al., 2003), represents the western- rectangular uniform-slip sub-faults to approximate bet-

Table 2 Parameters for the tsunamigenic strike-slip (grey background cells) and normal faults

Slip amount, length and width of all strike-slip faults are taken from Gu¨len et al. (2002). S. Tinti et al. / Marine Geology 225 (2006) 311–330 319 ter the along-fault variable strike. The slip varies in a the coastal node in the upper left corner of the mesh range of 1.0 m to 5.30 m, increasing from west to east. (Fig. 3), corresponding to the westernmost position on The dip-slip faults taken into account in our study the northern coast. The coastline is run clockwise, are the Kavakli fault (KAVF) and all the faults reaching the lower left corner. Relevant localities are corresponding to the step-over regions of the Karamu¨r- all within the I˙zmit bay and are identified with numb- sel (KARF) and of the I˙zmit (IF) sub-basins (see Table ers in the grid map given in Fig. 3 that are repeated in 2). The only inland fault is the Kavakli fault: for this Fig. 4 to help the reader. The results of the two fault, strike, length and slip values come from measure- simulations, with and without MF, are very similar, ments taken in the field (Barka et al., 2002; Lettis et al., with a slight difference that can be noted only for the 2002, and IT team observations), and there is clear coastal nodes that are found outside the˙ Izmit bay evidence that it ruptured during the Kocaeli earthquake. (plotted at the initial and ending part of the graph), The submarine normal faults are determined on the which means that the tsunamigenic effect of MF is not basis of bathymetric and shallow seismic surveys (S¸en- relevant. Therefore, it will not be considered any go¨r et al., 1999; Lettis et al., 2001, 2002), but there is longer in our analysis. The most important outcome no certainty that they were active during the main of the tsunami simulation is that the calculated MWE quake. Here we assume that all these faults have dip are very low along the entire coast of the bay with of 608 and are very shallow (fault width is supposed to values that are typically less than 30 cm and occa- be 3 km), being surface expression of the stress accom- sional peaks around 50–70 cm. This is at odd with the modation processes associated with the step-overs. We run-up height observations and is a very strong ele- assume further that they were involved in the main ment proving our thesis that the dextral fault system of shock process with a slip amount comparable with the the Kocaeli earthquake is not able to generate a sig- Kavakli fault (see Table 2). nificant tsunami. The initial tsunami energy is assumed to be only In the second stage of the analysis we invoke the potential with water elevation field taken to be equal to contribution of the normal faults given in Table 2. The the coseismic vertical displacement field of the seafloor, MWE corresponding to the case of the sole normal- which is usual in simulations of tsunamis induced by fault system and to the case of the ensemble of all the earthquakes. Surface coseismic deformations are calcu- active faults (strike-slip and normal) with the excep- lated through Okada’s (1992) formulas, holding for tion of MF are plotted in Fig. 4B. Furthermore, also rectangular faults buried in a homogeneous elastic the observed run-up heights are added to the graph. half-space. Since Okada’s model is based on elastic The two curves are found to differ insignificantly from linear theory, the displacements produced by multiple one another, which shows that the contribution of the rectangular sources can be superposed in order to obtain normal-fault sources is highly predominant in deter- the total coseismic ground displacement and, conse- mining the resulting MWE values. Most importantly, quently, the initial seawater elevation condition. the comparison between the calculated curves and the Our study intends to demonstrate that the tsunami measured run-up heights shows that the agreement is data cannot be explained only by the strike-slip fault acceptable, though not yet perfect. Though there are sources that were responsible for the catastrophic main some zones of marked overestimation (e.g. the peak ground shaking, and that additional sources are needed around point 20 corresponding to Defne Mahallesi), a in order to justify the observations. We suggest that general underestimation of the experimental data can these additional sources may be found in the set of be noted, which is however mild and can be probably coastal and submarine normal faults listed in Table 2. reduced or even eliminated by tuning the sources. The To this purpose we analyse the tsunamigenic effects of only town where the computed wave elevations are the strike-slip fault system first, and then we add the largely inconsistent with observations is Deðirmen- further contribution of the normal faults. As to the dere, where the tsunami has to be explained as the right-lateral fault tsunami, we run two different experi- effect of the coastal slump, as will be shown in detail ments, one with the sources involving the I˙zmit bay, in the next section. Fig. 5 depicts some of the tide- namely the three fault segments HF, GF and SF, and gauge records computed both on the northern and on the other with the supplementary addition of the Mar- the southern coast of the bay for the case of the mara fault (MF), whose role is not unanimously recog- combined normal and strike-slip faults. The curves nised. Fig. 4A exhibits the curves of the MWE are identified by numbers that correspond to the lo- computed along the coastal nodes of the basin for cality numbers shown in the map of Fig. 3 and also the strike-slip tsunami sources. The starting point is given in Table 1. Records either show a negative first 320 S. Tinti et al. / Marine Geology 225 (2006) 311–330

Fig. 4. Maximum water elevation computed for the coastal nodes calculated for the tsunamis induced by different set of sources. (A) All strike-slip sources either including the Marmara Fault (S+M) or excluding it (S). (B) All dip-slip faults (D) and dip-slip together with strike-slip faults (D+S). Faults are listed in Table 2 and graphed in Fig. 2. Distances on the horizontal axis are measured along the coast from the upper left corner of the grid given in Fig. 3 and numbers correspond to the localities on the same figure. Solid circles in panel (B) are the observed run-up heights given in Table 1. Maximum values reported in Deg˘irmendere (15 m) go out of scale. arrival or a very small positive arrival that is followed ple). In most places the largest oscillations occur in by a much larger sea level lowering, which is in the first 5–10 min after the shock with dominant agreement with most accounts of eyewitnesses who period of 3–4 min, and are later followed by a long observed a sea retreat first (the idea is that the initial train of small slow-decay oscillations also with longer sea rise was so small that passed unobserved to peo- period. The complex structure of the marigrams S. Tinti et al. / Marine Geology 225 (2006) 311–330 321

shoreline (Rathje et al., 2002, 2004; Cetin et al., 2004). A more violent tsunami was testified by witnesses and observed by post-event surveys in this area. It started at the same time of the main shock, or with some minutes delay according to only a single account, and manifested first with a seawater withdrawal (Altinok et al., 2001; Rothaus et al., 2004; IT team). In this work, based on data taken from the literature and on data collected by the IT team, first we analyse the stability of the mass involved in the failure, and then we simu- late its motion and compute the induced tsunami. Bathymetric data were collected by a boat survey undertaken around the coast of Deg˘irmendere on Oc- tober 1999, after the slide, by the IT team. We recon- struct the post-event bathymetry also with the aid of data extrapolated from the article by Cetin et al. (2004). The pre-event bathymetry is derived from the nautical map of the I˙zmit bay in conjunction with data from Cetin et al. (2004). Pre- and post-failure positions of the coastline are deduced also by making use of the maps by Rathje et al. (2002, 2004). We evaluated the volume of the slump by comparing pre- and post-event bathymetric data in the detachment area, obtaining the figure 6 3 Fig. 5. Computed tide-gauge records for the tsunami induced by all of 5.1810 m . faults, but the Marmara fault (case D+S of Fig. 4) for the localities Post-failure analyses of on-land cracks and geo- identified by the numbers on the right. technical investigations (borings, as well as in situ SPT, CPT and CPTU tests) carried out in the vicinity reflects the complexity of the sources and of the basin of the failed area are available at the web site http:// morphology, which favours the onset and persistence peer.berkeley.edu/turkey/adapazari (Cetin et al., 2004). of transverse mode of oscillations, corresponding to The analyses show the presence of distinct layers, the waves crossing the bay in north–south direction. Re- surface one (down to a 0.5–1 m depth) consisting of cord 24 corresponds to Topc¸ular that is found in the artificial fill of gravely sand and of silty clay. Deeper, outer I˙zmit bay basin and is far from the normal-fault one meets a sequence of thicker layers of silty sand, sources considered here. The tsunami is here quite sand and sandy silt with occasional presence of gravel, weak, which shows that the energy tends to remain the underlying bedrock being generally found deeper trapped within the inner basins of the bay with modest than 12–15 m (Cetin et al., 2004). outward radiation. Marigram 16 is computed at Deg˘irmendere. The main comment is that the calculated 6. Stability analysis oscillations here are small amplitude, and generally negative in the first post-earthquake 10-min window, The stability study of the Deg˘irmendere slump is during half of which the sea remains below the normal carried out here through an original method, hereafter level by about 1–1.5 m. This is a detail that may have designed as TM method, that is based on the limit some relevance for the stability of the local coastal equilibrium theory (Tinti and Manucci, in press), but slope, as will be discussed in the next section. that differs from the classical limit equilibrium methods. Since the method applied is new, we consider it 5. The slump at Deg˘irmendere useful for the reader to highlight briefly the main characteristics of this approach. The stability is ana- Deg˘irmendere is a village located in the southern lysed along 2D cross-sections. As in the classical limit coast of the Karamu¨rsel pull-apart basin (point 16 in equilibrium methods, the 2D section is partitioned into Fig. 3). A segment of coast about 300 m long and 75 m slices by means of vertical cuts. Imposing the condition wide slid into the sea carrying a multi-storey hotel and of the static equilibrium to all the slices results in a set two adjacent buildings, and produced a completely new of three equations (see Fig. 6): two for the horizontal 322 S. Tinti et al. / Marine Geology 225 (2006) 311–330 and vertical forces respectively and one for the torque. zontal stresses on the vertical downslope face of the In the present formulation they have the following slice (see Tinti and Manucci, in press, for all the form: details). The above set of three equations is comple- mented by the equation of limit equilibrium based on d E þ Ptana S Dtanb ¼k cosww the MohrCoulomb law: dx FS ¼ cV utanuV þ PtanuV d X þ P þ Stana D ¼ ðÞ1 þ k sinw w dx where cV(x) and /V(x) are material parameters (cohesion coefficient and friction angle respectively), while u(x) is the pore pressure, depending on the level of the d d z2 z1 A z1 E DtanbðÞ¼z2 z1 k cosww piezometric line. All these functions are allowed to dx dx 2 vary downslope, i.e. are supposedly depending on the Here the coordinate x runs horizontally in the down- variable x. The quantity F is the factor of safety, which slope direction. The slide cross-section is bounded by discriminates between a state that for the mass is un- b N the bottom profile z1(x) and by the top profile z2(x), and stable ( F 1) and one that is stable ( F 1). is assumed to be of unitary transverse width. The local The boundary conditions to impose at the up-hill and slopes of these profiles are denoted respectively by a(x) frontal ends of the slide, with respective coordinates xi and by b(x). The quantities appearing in the equations and xf, concern the components E and X of the inter- are: the pressure P and the shear stress S, acting at the slice forces and the torque A: base of the slice respectively in the normal and tangen- tial direction; the hydrostatic load D that is supposedly ExðÞ¼i ExðÞ¼f 0 acting at the top of the slice and is normal to it; the inter-slice force per unit width, decomposed into the XxðÞ¼XxðÞ¼0 horizontal and vertical components E and X; the slide i f weight per unit horizontal length and unit width w; and the seismic load, acting at the centre of mass of the slide AxðÞ¼i AxðÞ¼f 0 along the direction determined by the angle w (that grows clockwise from the horizontal axis oriented The above system of equations counts as many as downslope as is shown in Fig. 6) and with magnitude five unknown functions, namely P(x), S(x), E(x), X(x) proportional to the slice weight through the coefficient and A(x), and one unknown parameter, i.e. the safety k. The variable A in the torque balance equation has the factor F. It is underdetermined and admits more than a dimension of a torque per unit width and is defined as single solution, since the number of unknowns exceeds the first-order moment of the distribution of the hori- the number of constraining equations. Traditional limit equilibrium methods, such as Bishop’s, Janbu’s, Spen- cer’s, Morgenstern and Price’s, etc., seem to overlook the problem, under the assumption that what matters is the value of the factor of safety F, and non-uniqueness has only a minor effect on it. In other words, it is traditionally believed that though many (virtually, an infinite number of) solutions can be found, the resulting value of F varies in an acceptably narrow range. We were able to show that this opinion is not well founded, since exact solutions can be found that correspond to a very large range of F, both below and beyond the critical value of 1. Consequently, this means that usu- ally there exist solutions showing that the slope is unstable as well as solutions pointing conversely to a slope stability. In order to overcome this basic draw- Fig. 6. Sketch of the vertical cross-section of the sliding body cut into slices, with representation of the forces acting on each single slice. All back, we developed a new approach by introducing a variables are defined in the text. Here W is the total weight of the slice new parameter we call the lithostatic deviation d, which and is given by the product of c times the slice horizontal length. is expressed as the dimensionless ratio of the average S. Tinti et al. / Marine Geology 225 (2006) 311–330 323

all the five equations given above together with the six boundary conditions and that, in addition, minimises the lithostatic deviation d. The new approach changes drastically the way the solution can be searched for. The stability factor is not treated as an unknown parameter anymore, but it is taken as a given parameter that can vary within a specified range including the critical value 1. The method we use consists in expressing the unknown function X into a truncated Fourier expansion, and then to find the ensemble of exact solutions to the stability problem corresponding to any pre-given value of F. For each solution, the lithostatic deviation is computed, and, according to the MLD criterion, one picks up the value of F that is associated with the solution corresponding to the minimum value of d (Tinti and Manucci, in press). To study the Deg˘irmendere slump the TM method was applied together with various traditional limit equi- Fig. 7. Reconstruction of the pre- and post-slump coastline (Rathje et librium methods. We selected two profiles along the al., 2002, 2004; Cetin et al., 2004). The footprint of the failing mass body that are aligned with the direction of sliding. (dashed line) and profiles A and B (solid lines) of the analysed These are given in Fig. 7 where the reconstructed vertical cross-sections are also shown. footprint of the failing mass is portrayed. The two cross-sections, denoted by A and B, are taken along magnitude of the inter-slice forces to the weight of the the average direction of sliding motion and are given in slide: Fig. 8: the top curve has an average slope of about 158. Z To perform the stability analysis the bottom boundary is xf 1=2 1 2 2 approximated by an arc-like curve. Notice that this is d ¼ W 1 ExðÞ þ XxðÞ dx ðÞx x not a requirement imposed by our approach, for it can Z f i xi xf be applied to slip boundaries with arbitrary shape, but W ¼ wdx rather it is a convention adopted by the traditional x i methods, and we adhere to it to allow comparison. and by further adopting the principle of Minimum We have studied the stability of the body in three Lithostatic Deviation (MLD). It asserts that the state different situations: in the pre-earthquake condition, of limit equilibrium of the body is the state that satisfies during the shock, and also in the course of the first

Fig. 8. Reconstructed vertical cross-sections of the slump body taken along transects A and B shown in Fig. 7. The assumed basal slip surface is circular and plotted as a dashed line. In the stability analysis the material is taken to be cohesionless (cV=0) and with angle of friction /V=298. 324 S. Tinti et al. / Marine Geology 225 (2006) 311–330

Table 3 Safety factor F and lithostatic deviation d computed for the two body profiles A and B shown in Fig. 8 in the absence and in the presence of seismic load Profile A, earthquake: k =0.45, Profile B, earthquake: k =0.45, Profile A, no earthquake, Profile B, no earthquake, w =228, tsunami: sl=1m w =218, tsunami: sl=1m no tsunami no tsunami F d F d F d F d Ordinary 0.848 – 0.951 – 5.842 – 7.061 – Bishop 0.903 0.1591 0.986 0.1578 5.963 0.0685 7.128 0.0528 Janbu 0.884 0.1559 0.989 0.1560 5.884 0.0681 7.076 0.0526 Spencer 0.908 0.1738 0.989 0.1714 5.963 0.0680 7.127 0.0525 Morg and 0.908 0.1746 0.989 0.1724 5.963 0.0679 7.127 0.0525 Price Our method 0.906 0.1620 0.989 0.1639 5.966 0.0672 7.115 0.0523 Observe that classical limit equilibrium methods and our approach provide quite similar results for this slope. The body is found to be very stable in pre-earthquake conditions (k =0), which is mainly due to the bottom friction and to the stabilising hydrostatic load of the sea. Conversely, F collapses to critical values of instability in case of seismic load and tsunami attack: the seismic acceleration is taken as large as 0.45 g and the sea is assumed to be 1 m below the ordinary level, in consequence of the water retreat caused by the arriving tsunami. The value used for w is the one that is found to produce the maximum instability. Notice that the methods given in the first three rows do not provide exact solutions to the stability equations, while Spencer’s, Morgenstern and Price’s and ours do. arrival phase of the earthquake-induced tsunami. The hydrostatic load exerted by the weight of the seawater results are summarised in Table 3 and in Fig. 9. Table 3 column. In the problem formulation highlighted above, concerns both profiles A and B, and provides safety the seismic load is introduced by means of a static force factors computed by applying traditional methods as that is applied to the centre of mass of each slice along well as our approach. The first observation is that the the direction specified by the seismic angle w that is results happen to be quite similar, which helps us focus measured clockwise from the horizontal axis. Conse- on the physics of the problem more than on algorithms. quently, w =908 identifies a seismic force pointing ver- The pre-earthquake condition is a condition with no tically downward and w =08 specifies a downslope seismic load (k =0) and with normal sea level. The horizontal force (see the sketch of Fig. 6). The magni- coastal slope results to be totally stable owing to the tude of the force is given through the corresponding concurrent effects of the bottom friction and of the acceleration that is scaled to the gravity acceleration g

Fig. 9. Factor of safety vs. the orientation of the seismic force kW, expressed through the angle w defined in Fig. 6. S. Tinti et al. / Marine Geology 225 (2006) 311–330 325 by means of the factor k. In our computations we have term cannot be neglected in computing F, since this assumed values for k as large as 0.45 and 0.55, which would lead to unacceptable underestimation. Observe are quite reasonable in view of the experimental data set that varying the water level by 1 m implies differences of recorded peak ground accelerations (Barka et al., in the order of 0.05 in the resulting values of the factor 2002; Cetin et al., 2004) and of acceleration modeling of safety. These are not large, but are sufficient to (Yagi and Kikuchi, 2000). The first arrival of the tsu- support the idea that the tsunami may have contributed nami that was induced by the earthquake caused the sea to destabilising the slope, though it could be hardly to retreat from the usual shoreline almost along the seen as the main cause of the failure. entire coast of the I˙zmit bay, and Deg˘irmendere was no exception. Hence, as regards the stability analysis, 7. Simulations of the slump and of the induced the condition of the tsunami attack is represented by tsunami lowering the sea level by 1 m, which implies a reduction of the hydrostatic load on the submarine slope. Tsunami simulations showed a satisfactory agree- Columns 1 and 2 of Table 3 show results of compu- ment with real data, assuming the whole set of the tations with both loads applied: calculations refer to the strike- and dip-slip faults as the triggering cause, except value of w that is found to be the one producing the in the Deg˘irmendere area, where observations are great- most unstable condition for a given profile. The depen- ly underestimated. In this section we investigate the dence of F from the angle w is explored in detail in Fig. motion of the slump and the characteristics of the 9, where all computations are performed by means of generated tsunami, by means of a numerical simulation our approach on the profile B. The various curves method that is based on a double simulation code, one plotted here serve to show the effect of changing the for the slump and one for the tsunami. hydrostatic and the seismic load on the resulting factor The computation of the slump dynamics makes use of safety. All curves are quite regular and display a well of a numerical Lagrangian model that divides the slump defined minimum, corresponding to negative values of into a 2D matrix of deformable blocks (24 in the w, and show that the most destabilising seismic load present case) with quadrilateral basis and calculates has both a downslope and an upward component. the motion of each block taking into account all the Increasing the seismic force magnitude from k =0.45 relevant forces, namely gravity, buoyancy, the basal to k =0.55 reduces the safety factor by 0.1–0.2 and surface friction, the seawater drag on the frontal and widens sensibly the interval of angles that may give upper body surfaces, as well as the interaction with the rise to instability. Conversely, increasing the hydrostatic next neighbours (see Tinti et al., in press for a more load has the effect of making the slope more stable. detailed description). Fig. 10 displays the curves of the This is clearly seen by comparing the three curves absolute value of the computed block velocities vs. corresponding to the same seismic load (k =0.45), but time, as well as the speed of the slump centre of mass with different assumed value of water level. The case of that is obtained as the weighted average of the blocks no sea load is added in the figure only to prove that this speed. Blocks move with rather different velocities in

Fig. 10. Absolute value of the velocities of the blocks forming the slump vs. time. The mean velocity is plotted as a continuous black line. 326 S. Tinti et al. / Marine Geology 225 (2006) 311–330

Fig. 11. 3D view of the trajectories of the 24 blocks forming the slide computed by means of the numerical model. the initial phase of strong acceleration, and then, during The induced tsunami is simulated through a shallow- the subsequent deceleration phase, they tend to move water code based on the finite-element technique that more uniformly, with the consequence that the entire solves the non-linear Navier-Stokes equations for invis- mass advances almost like a rigid body. The slide is cid incompressible fluid. The model accounts for the rather slow, attaining the maximum speed of about 9 m/ wave forcing due to the mass movement by means of an s, and comes to a rest in slightly more than 2 min. excitation term that depends on the rate of vertical Consequently, the computed run-out is short, about 500 displacement of the sea bottom Bth, modified through m. The block trajectories, depicted in the 3D perspec- a suitable filtering function depending on the local tive of Fig. 11, run along the maximum bathymetric depth and on the lateral size of the slide. Here gradient (first northwestward and then westward), end- h(x,y,t)=H(x,y)+Hs(x,y,t) represents the sea bottom ing at about 45–50 m depth. as the sum of the time-independent bathymetry function

Fig. 12. Finer grid used in the slump-induced tsunami simulation. Notice that the grid has very high resolution in the Deg˘irmendere-Tupras region where the tsunami is generated and propagates first. S. Tinti et al. / Marine Geology 225 (2006) 311–330 327

Fig. 13. Snapshots of the calculated water elevation fields in the source area at different times after the slide initiation.

Fig. 14. Water maxima and minima along the coast. Observe the fast decay of the tsunami away from the source: f.i. see that in the I˙zmit area the tsunami is very weak. The distance is measured along the coast starting from the node that is in the northwestern corner of the grid of Fig. 12.We remark that the vertical peaks computed in Deg˘irmendere go out of scale and exceed 8 m. 328 S. Tinti et al. / Marine Geology 225 (2006) 311–330

H(x,y) and the landslide term Hs(x,y,t), which is the and short-period, with a clear sea retreat occurring first. instantaneous thickness of the body seen in the point The largest disastrous wave series, composed of a few (x,y) at the time t. Further details on the model and on (two–three) waves, attacks the Deðirmendere coast the filtering function are omitted here and may be found within the first minute and the wave energy is concen- in previous papers addressing analogous problems of trated in a short time window less than 100 s. From Fig. tsunamis induced by landslides (see Tinti et al., 1999, in 13 one learns that the tsunami arrives at the opposite press). The finite-element grid, consisting of 7451 coast in less than 2 min, and from the marigrams of Fig. nodes and 13,563 triangles, is built with high-resolution 15 one sees that Tupras is reached soon after by longer in the source region where the typical element size is period waves with a positive leading front. less than 20 m, and covers the central and eastern part of the I˙zmit bay, as may be seen from Fig. 12. Snap- 8. Conclusions shots of the tsunami wave field are portrayed in Fig. 13, from which we can notice how the wave propagates The study conducted in this paper has shown that the from Deg˘irmendere and, after reaching the opposite tsunami that was observed in the I˙zmit bay in the night coast, travels eastward (slower) and westward (faster) of August 17, 1999 can be explained only as the with different speeds owing to the different water depth combined effect of several sources: tectonic coseismic of the basins. Fig. 14 displays the maxima and the displacements and local mass failure. One relevant minima of the water elevation computed in the coastal result is that the sea bottom displacements computed nodes. It can be seen that the tsunami is very strong at by means of the classical elastic models, such as Oka- the source with a peak–peak height exceeding 12 m, but da’s, from the set of the five right-lateral faults, that it drops drastically as soon as one moves some kilo- were the chief responsible for the disastrous seismic meters away. At Tupras, that is found on the opposite waves (see Table 2), are not sufficient to produce a side of the bay, the maximum amplitude of the wave is sizeable tsunami. And this irrespective of the fact that around 3 m, and less for the other towns. Synthetic tide- the main rupture zone of the Kocaeli earthquake is or is gauge records for some grid nodes corresponding to not assumed to extend with an additional fault into the Deg˘irmendere and Tupras are shown in Fig. 15. Ob- eastern part of the Marmara Sea, since the tsunamigenic serve that in the source the waves are high-amplitude contribution of the Marmara fault is negligible. In order to explain the main features of the regional tsunami affecting the I˙zmit bay, dip-slip faults have to be invoked capable to produce significant offset of the seafloor in the bay. In this paper we have assumed that such faults may be found in the system of normal faults that are associated with the step-over zones marking the transition from one strike-slip fault segment to the next (see Fig. 2). In particular, we have considered the graben-like fault systems in the central (Karamu¨rsel) and eastern (I˙zmit) basins of the bay. The selected fault parameters match the idea that these are very shallow ancillary faults that contributed only marginally to the catastrophic seismic shaking of the main earthquake. They are however important enough to induce relevant vertical movement in the sea bottom. The computer simulation of the tsunami shows that only with the inclusion of these additional sources the run-up heights that were observed or assessed by the post-tsunami field surveys and that fall in the interval 1–3.5 m can be reproduced reasonably. The exception of Deg˘irmendere that was attacked by waves several meters high according both to eyewitnesses and to post-event observations is due to a local slump. The stability analysis carried out Fig. 15. Time histories computed in coastal nodes in the source region on the basis of geotechnical data (defining the material (Deg˘irmendere) and in Tupras on the opposite coast of the bay. property of the failing mass) and of bathymetric surveys S. Tinti et al. / Marine Geology 225 (2006) 311–330 329

(capable of detailing the sliding surface and the run-out Ambraseys, N.N., Finkel, C.F., 1991. Long term seismicity of I˙stan- distance) shows that the body was mainly destabilised bul and of the Marmara Sea region. Terra Nova 3, 527–539. Ambraseys, N.N., Finkel, C.F., 1995. The Seismicity of Turkey and by the seismic load, but it also suggests that the tsunami Adjacent Areas. A Historical Review 1500–1800. Eren Yayinlari, may have had a triggering role. In fact the first mani- I˙stanbul. 240 pp. festation of the regional tsunami was here, like in most Arel, E., Kiper, B., 2000. The coastal landslide occurred by August other coastal segments, a sea withdrawal, and a sea 17, 1999 earthquake at Degirmendere (Kocaeli). Proceedings of level lowering with the corresponding decrease of hy- Coastal Engineering 3rd National Symposium, October 5–7, 2000. Ç anakkale, Turkey, pp. 45–55. (in Turkish). drostatic load on the slope was shown to have the effect Armijo, R., Meyer, B., Hubert, A., Barka, A., 1999. Westward of decreasing the safety factor of the slide. The simu- propagation of the North Anatolian fault into the Northern Aege- lation of the slump motion was carried out by a suitable an: timing and kinematics. Geology 27, 267–270. landslide evolution model that shows that the slump Armijo, R., Meyer, B., Navarro, S., King, G., Barka, A., 2002. moves rather slow, attaining speed at most less than 10 Asymmetric slip partitioning in the Sea of Marmara pull-apart: a clue to propagation processes of the North Anatolian Fault? Terra m/s. The outcome of the slide model was used to Nova 14, 80–86. compute the forcing term in the hydrodynamic equa- Barka, A., 1992. The North Anatolian fault zone. Ann. Tecton. VI, tions governing the tsunami generation and propaga- 164–195. (Suppl.). tion. Computer tsunami simulations via a finite-element Barka, A., Kadinsky-Cade, K., 1988. Strike-slip fault geometry in model have shown that the slump generation mecha- Turkey and its influence on earthquake activity. Tectonics 7, 663–684. nism is capable of producing very large short-period Barka, A., Akyu¨z, H.S., Cohen, H.A., Watchorn, F., 2000. Tectonic waves that are very local and match the tsunami obser- evolution of the Niksar and Tasova-Erbaa pull-apart basins, North vations in Deg˘irmendere. Anatolian Fault Zone: their significance for the motion of the Anatolian block. Tectonophysics 322, 243–264. Acknowledgements Barka, A., Akyu¨z, H.S., Altunel, E., Sunal, G., Ç akir, Z., Dikbas, A., Yerli, B., Armijo, R., Meyer, B., Chabaler, J.B, Rockwell, T., Hartleb, J., Dawson, T., Christofferson, S., Tucker, A., Fumal, T., The research has been funded partly by INGV (Isti- Langridge, R., Stenner, H., Lettis, W., Bachhuber, J., Page, W., tuto Nazionale di Geofisica e Vulcanologia, Rome, 2002. The surface rupture and slip distribution of the 17 August Italy) and partly by MIUR (Ministero dell’Istruzione, 1999 I˙zmit earthquake (M 7.4) North Anatolian Fault. Bull. dell’Universita` e della Ricerca) to study tsunami gen- Seismol. Soc. Am. 92, 43–60. Ç akir, Z., de Chabalier, J.B., Armijo, R., Meyer, B., Barka, A., erated by earthquakes and by underwater mass move- Peltzer, G., 2003. Coseismic and early post-seismic slip associated ments. This work was partly supported by the Research with the 1999 I˙zmit earthquake (Turkey), from SAR interferom- Fund University under the Project Number: UDP-328/ etry and tectonic field observations. Geophys. J. Int. 155, 93–110. 03062004 allocated to Y. Altinok. Cetin, K.O., Isik, N., Unutmaz, B., 2004. Seismically induced landslide at Deg˘irmendere Nose, I˙zmit bay during Kocaeli (I˙zmit)- Turkey earthquake. Soil Dyn. Earthqu. Eng. 24, 189–197. References Delouis, B., Giardini, D., Lundgren, P., Salichon, J., 2002. Joint inversion of InSAR, GPS, teleseismic, and strong-motion data Alpar, B., 1999. Underwater signatures of the Kocaeli earthquake for the spatial and temporal distribution of earthquake slip: Ap- (August 17th 1999). Turk. J. Mar. Sci. 5, 111–130. plication to the 1999 I˙zmit mainshock. Bull. Seismol. Soc. Am. Alpar, B., Yaltirak, C., 2002. Characteristic features of the North 92, 278–299. Anatolian Fault in the eastern Marmara Region and its tectonic Feigl, K.L., Sarti, F., Vadon, H., McClusky, S., Ergintav, S., Durand, evolution. Mar. Geol. 190, 329–350. P., Bu¨rgmann, R., Rigo, A., Massonet, D., Reilinger, R., 2002. Altinok, Y., Ersoy, S¸., 2000. Tsunamis observed on and near the Estimating slip distribution for the I˙zmit mainshock from coseis- Turkish coast. Nat. Hazards 21, 185–205. mic GPS, ERS-1, RADARSAT, and SPOT measurements. Bull. Altinok, Y., Alpar, B., Ersoy, S¸., Yalçiner, A.C., 1999. Tsunami Seismol. Soc. Am. 92, 138–160. generation of the Kocaeli Earthquake (August 17th 1999) in the Flerit, F., Armijo, R., King, G.C.P., Meyer, B., Barka, A., 2003. Slip I˙zmit Bay: coastal observations, bathymetry and seismic data. partitioning in the Sea of Marmara pull-apart determined from Turk. J. Mar. Sci. 5, 131–148. GPS velocity vectors. Geophys. J. Int. 154, 1–7. Altinok, Y., Tinti, S., Alpar, B., Yalçiner, A.C., Ersoy, S¸., Bortolucci, Gu¨len, L., Pinar, A., Kalafat, D., Ozel, N., Horasan, G., Yilmazer, M., E., Armigliato, A., 2001. The Tsunami of August 17, 1999 in Isikara, A.M., 2002. Surface fault breaks, aftershocks distribution, I˙zmit Bay, Turkey. Nat. Hazards 24, 133–146. and rupture process of the 17 August 1999 I˙zmit, Turkey, earth- Ambraseys, N.N., 1962. Data for the investigation of the seismic sea- quake. Bull. Seismol. Soc. Am. 92, 230–244. waves in the Eastern Mediterranean. Bull. Seismol. Soc. Am. 52, Gu¨neysu, A.C., 1999. The bathymetry of the I˙zmit bay. Turkish. J. 895–913. Mar. Sci. 5, 167–169. Ambraseys, N.N., 2002a. The seismic activity of the Marmara Sea Hubert-Ferrari, A., Barka, A., Jacques, E., Nalbant, S.S., Meyer, B., region in the last 2000 years. Bull. Seismol. Soc. Am. 92, 1–18. Armijo, R., Tapponier, P., King, G.C.P., 2000. Seismic hazard in Ambraseys, N.N., 2002b. Seismic sea-waves in the Marmara Sea the Marmara Sea region following the 17 August 1999 I˙zmit region during the last 20 centuries. J. Seismol. 6, 571–578. earthquake. Nature 404, 269–273. 330 S. Tinti et al. / Marine Geology 225 (2006) 311–330

Hubert-Ferrari, A., Armijo, R., King, G., Meyer, B., Barka, A., 2002. Sucuog˘lu, H., Gu¨lkan, P., BakVr, B.S., O¨ zcebe, G., Tankut, T., YVlmaz, Morphology, displacement, and slip rates along the North Anato- Ç ., Ersoy, U., Erberik, A., Akkar, S., Gu¨r, T., 2000. Kocaeli and lian Fault, Turkey. J. Geophys. Res. 107 (B10), 2235. doi:10. Du¨zce earthquakes: Engineering Report, Joint report of the Turk- 1029/2001JB000393. ish Chamber of Civil Engineers and the Middle East Technical Kuran, U., Yalçiner, A.C., 1993. Crack propagations earthquakes and University, Ankara, 174 pp., (in Turkish). tsunamis in the vicinity of Anatolia. In: Tinti, S. (Ed.), Tsunamis Tinti, S., Armigliato, A., 2003. The use of scenarios to evaluate in the World. Kluwer Academic Publishers, pp. 159–175. tsunami impact in South Italy. Mar. Geol. 199, 221–243. Kus¸çu, I˙., Okamura, M., Matsuoka, H., Awata, Y., 2002. Active faults Tinti, S., Bortolucci, E., 1999. Strategies of optimal grid generation in the gulf of I˙zmit on the North Anatolian fault, NW Turkey: a for finite-element tsunami models. Proc. International Conference high resolution shallow seismic study. Mar. Geol. 190, 421–443. on Tsunamis, Paris, 26–28 May 1998, pp. 269–294. Le Pichon, X., S¸engo¨r, A.M.C., Demirbag˘, E., Ranging, C., I˙mren, C., Tinti, S., Manucci, A., in press. Gravitational stability computed Armijo, R., Go¨ru¨r, N., Çag˘atay, N., Mercier de Lepinay, B., through the limit equilibrium method revisited. Geophys. J. Int. Meyer, B., SaatçVlar, R., Tok, B., 2001. The active main Marmara Tinti, S., Gavagni, I., Piatanesi, A., 1994. A finite-element numerical fault. Earth Planet. Sci. Lett. 192, 595–616. approach for modelling tsunamis. Ann. Geophys. 37, 1009–1026. Lettis, W., Bachhuber, J., Barka, A., Witter, R., Brankman, C., Tinti, S., Bortolucci, E., Armigliato, A., 1999. Numerical simulation 2001. Surface fault rupture and segmentation during the Kocaeli of the landslide-induced tsunami of 1988 in Vulcano island, Italy. earthquake. In: Barka, A., Kozaci, I˙., Akyu¨z, S., Altunel, E. Bull. Volcanol. 61, 121–137. (Eds.), The 1999 I˙zmit and Du¨zce earthquakes: Preliminary Tinti, S., Pagnoni, G., Zaniboni, F., in press. The landslides and Results. I˙stanbul Technical University, I˙stanbul, Turkey, ISBN: tsunamis of 30th December 2002 in Stromboli analysed through 975-561-182-7 pp. 31–54. numerical simulations. Bull. Volcanol. Lettis, W., Bachhuber, J., Witter, R., Brankman, C., Randolph, C.E., US Geological Survey, 2000. Implications for earthquake risk reduc- Barka, A., Page, W.D., Kaya, A., 2002. Influence of releasing tion in the United States from the Kokaeli, Turkey, earthquake of step-overs on surface fault rupture and fault segmentation: exam- August 17, 1999. USGS Circular 1193, p. 65. Available online at: ples from the 17 August 1999 I˙zmit earthquake on the North http://geology.cr.usgs.gov/pub/circulars/c1193/index.html. Anatolian Fault, Turkey. Bull. Seismol. Soc. Am. 92, 19–42. Yagi, Y., Kikuchi, M., 2000. Source rupture process of the Kocaeli, Okada, Y., 1992. Internal deformation due to shear and tensile faults Turkey, earthquake of August 17, 1999, obtained by joint inver- in a half-space. Bull. Seismol. Soc. Am. 82, 1018–1040. sion of near-field data and teleseismic data. Geophys. Res. Lett. Papadopoulos, G.A., 2001. Tsunamis in the Eastern Mediterranean: a 27, 1969–1972. catalogue for the area of Greece and adjacent seas. Proceedings of Yalçiner, A.C., Altinok, Y., Synolakis, C.E., 2000. Tsunami Waves in the Joint IOC–IUGG International Workshop: Tsunami Risk As- I˙zmit Bay after the Kocaeli Earthquake. Earthquake Spectra, sessment Beyond 2000: Theory, Practice and Plans, Moscow, Kocaeli, Turkey Earthquake of August 17, 1999 Reconnaissance Russia, June 14–16, 2000, pp. 34–43. Report, Special Issue Supplement to Professional Journal of Parsons, T., 2004. Recalculated probability of M z7 earthquakes Earthquake Engineering Research Institute vol. 16, pp. 55–62. beneath the Sea of Marmara, Turkey. J. Geophys. Res. 109 Yalçiner, A.C., Imamura, F., Kuran, U., Ç akirog˘lu, Y., O¨ zbay, I., (B05304). doi:10.1029/2003JB002667. Erdinç, E., Durmus¸og˘lu, O¨ ., Altinok, Y., Alpar, B., 2001a. An Parsons, T., Toda, S., Stein, R.S., Barka, A., Dieterich, J., 2000. integrated model on the generation and propagation of tsunamis in Heightened odds of large earthquakes near I˙stanbul: an interac- the surrounding seas around Anatolia, Tu¨rkiye çevresi denizlerde tion-based probability calculation. Science 288, 661–665. tsunami dalgasi hareketi için biles¸ik model olus¸turulmasi, . Final Piatanesi, A., Tinti, S., 1998. A revision of the 1693 eastern Sicily Report of Basic Research Project No: TUBITAK-YDABÇ AG-60, earthquake and tsunami. J. Geophys. Res. 103, 2749–2758. November 2001, pp. 147 (in Turkish). Piatanesi, A., Tinti, S., 2002. Numerical modeling of the September 8, Yalçiner, A.C., Kuran, U., Altinok, Y., Alpar, B., Ersoy, S¸., 2001b. 1905 Calabrian (southern Italy) tsunami. Geophys. J. Int. 150, Historical earthquakes and associated tsunamis in the Aegean 271–284. Sea, Tu¨rkiye kiyilarinda tarihsel depremler ve onlara ilis¸kin Rathje, E.M., Wright, S.G., Bachhuber, J., 2002. Evaluation of inte- tsunamlerin izlerinin aras¸tirilmasi. Final Report of Basic Re- grated seismic hazards and ground failure in pull-apart basin search Project No: TUBITAK-INTAG-827, September 2001, during 1999 Kocaeli earthquake, Turkey. Report to United States pp. 74 (in Turkish). Geological Survey, Earthquake Hazard Reduction Program (avail- Yalçiner, A.C., Synolakis, C.E., Alpar, B., Borrero, J., Altinok, Y., able online at: http://erp-web.er.usgs.gov/reports/annsum/vol43/ Imamura, F., Tinti, S., Ersoy, S¸., Kuran, U., Pamukçu, S., Kano- pt/g0043.pdf). g˘lu, U., 2001c. Field surveys and modeling of the 1999 I˙zmit Rathje, E.M., Karatas, I., Wright, S.G., Bachhuber, J., 2004. Coastal Tsunami. Proceedings of the International Tsunami Symposium, failures during the 1999 Kocaeli earthquake in Turkey. Soil Dyn. August 7–10, 2001, Seattle, Washington, USA, pp. 557–564. Earthqu. Eng. 24, 699–712. Yalçiner, A.C., Alpar, B., Altinok, Y., O¨ zbay, I˙., Imamura, F., 2002. Rothaus, R., Reinhardt, E., Noller, J., 2004. Regional considerations Tsunami in the sea of Marmara, historical documents for the past, of coastline change, tsunami damage and recovery along the models for the future. Mar. Geol. 190, 445–463. southern coast of the Bay of I˙zmit (The Kocaeli (Turkey) earth- Yalçiner, A.C., Pelinovsky, E., Talipova, T., Kurkin, A., Kozelkov, A., quake of 17 August 1999). Nat. Hazards 31, 233–252. Zaitsev, A., 2004. Tsunamis in the Black sea: comparison of the S¸engo¨r, A.M.C., Demirbag˘, E., Tuÿsu¨z, E., Kurt, H., Go¨ru¨r, N., historical, instrumental and numerical data. J. Geophys. Res. 109, Kus¸çu,˙ I., 1999. A preliminary note on the structure of the Gulf C12023. doi:10.1029/2003JC002113. of I˙zmit: implications for the westerly prolongation of the North Yu¨ksel, Y., Alpar, B., Yalçiner, A.C., Ç evik, E., Ozgu¨ven, O., Anatolian fault. In: Karaka, M., Ural, D. (Eds.), ITU-IHAS Inter- Ç elikog˘lu, Y., 2002. Effects of the eastern Marmara Earthquake national Conference on the Kocaeli Earthquake, 17 August 1999, on marine structures and coastal areas. Water Marit. Eng. 156, ITU Publication, pp. 25–38. 147–163.